Joint optimization under hyperparameter interdependencies in CIM-CACm

Establish a methodology to identify and jointly optimize sets of interdependent hyperparameters in the Coherent Ising Machine employing Chaotic Amplitude Control with momentum (CIM-CACm) when multiple hyperparameters simultaneously affect solution quality, rather than optimizing each parameter independently.

Background

The paper proposes an algorithm portfolio approach for hyperparameter tuning in Coherent Ising Machines using the CACm algorithm. The proposed Method A and Method B reduce the search space by optimizing one parameter at a time and, for Method B, prioritize parameters based on an initial evaluation.

While this sequential strategy improves performance over simultaneous high-dimensional Bayesian optimization, the authors note a limitation: real-world performance can depend on interactions among multiple hyperparameters. Ignoring such interdependencies may hamper search efficiency and result quality, especially for algorithms like CMA-ES that leverage covariance among parameters.